Server hardware fault early warning system and method based on recurrent neural network

Abstract

The invention belongs to the technical field of computer fault diagnosis, and particularly relates to a server hardware fault early warning system and method based on a recurrent neural network, and the method comprises the steps: constructing a server hardware fault early warning model; the method comprises the following steps: collecting time sequence working data of each key component of a server in real time, performing data preprocessing on the collected time sequence working data, and inputting the data into a server hardware fault early warning model for training; extracting fault features of the time sequence working data of each component; initializing weights and thresholds of fault features of the time sequence working data of each component extracted by the server hardware fault early warning model; optimizing the weight and the threshold value of the fault feature of the time sequence working data of each component; and inputting the real-time working data to the server hardware fault early warning model, and outputting a server hardware fault prediction result in real time. The server fault prediction method provides more accurate server fault prediction capability, has dynamic adaptive capability, can dynamically adjust related parameters, and enhances flexibility and practicability.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer fault diagnosis, and in particular to a server hardware fault early warning system and method based on a recurrent neural network. Background Art

[0002] Server hardware failure refers to a substantial failure of the server system hardware, such as components, integrated circuits, etc. Hardware failure is a common and serious problem in large-scale data centers and enterprise-level server environments. Hardware failure not only causes server downtime and data loss, but also causes production losses and increased maintenance costs.

[0003] Existing server hardware failure early warning systems often have the following challenges and limitations: First, many early warning systems rely on static rules and preset thresholds to judge hardware failures, and lack adaptability to dynamic changes in the system. Second, some early warning systems use simple algorithms, such as threshold judgment or rule engines, which cannot effectively process complex failure modes and time series data. Third, the time series data of server operating parameters usually has complex characteristics, including periodic changes, trend changes, and abnormal bursts, which require more complex analysis methods to process. Fourth, the accuracy and reliability of the early warning system directly affect the timely detection and handling of faults. Simple algorithms and rules may lead to a high false alarm rate or missed reports.

[0004] Existing early warning systems for server hardware failures include static rule systems based on thresholds: many early warning systems use fixed thresholds to monitor server operating parameters, and trigger early warnings once the set thresholds are exceeded. This method is simple and intuitive, but lacks flexibility and adaptability, and cannot cope with complex failure modes.

[0005] Prediction system based on statistical model: Statistical models and time series analysis methods are used to predict the changing trends and abnormal conditions of server working parameters. Although these methods can improve the accuracy of prediction to a certain extent, they are limited in their ability to handle complex time series data and abnormal patterns.

[0006] Therefore, the existing server hardware fault warning system has a high false alarm rate or missed alarms, lacks flexibility and adaptability, is unable to cope with complex failure modes, and has a low accuracy rate, which is a technical problem that urgently needs to be solved. Summary of the invention

[0007] The purpose of the present invention is to provide a software fault pre-diagnosis method and system by mounting a domestic server, so as to predict software faults before they occur, improve the efficiency of server software diagnosis, and implement software and server maintenance processing before software faults occur, thereby reducing the probability of software faults occurring.

[0008] In a first aspect, a server hardware failure early warning method based on a recurrent neural network is provided, comprising the following steps:

[0009] S1: Build a server hardware failure warning model;

[0010] S2: collect the time series work data of each key component of the server in real time, pre-process the collected time series work data, and input it into the server hardware fault warning model for training;

[0011] S3: Extract the fault characteristics of the time series working data of each component through the server hardware fault early warning model;

[0012] S4: Initialize weights and thresholds for the fault features of the time series working data of each component extracted by the server hardware fault early warning model;

[0013] S5: Optimizing the weights and thresholds of the fault characteristics of the sequential working data of each component;

[0014] S6: Input real-time working data into the server hardware failure warning model, and output the server hardware failure prediction results in real time.

[0015] Preferably, the server hardware failure warning model constructed in step S1 includes an input layer, a hidden layer and an output layer, and its given sequence input learning data is:

[0016] X={X1+X2+···+X t};

[0017] Among them, X1, X2, X t is the time series data, t is the expansion length of the input learning data;

[0018] The formula for a cycle unit at time t is:

[0019] S t =f(W*S (t-1) , U*X t );

[0020] Among them, S t is the hidden layer, t is the cycle unit time; S (t-1) The current time S t The hidden layer of the previous time step, S t Need to use the S of the previous time step (t-1) , f is the encapsulated feedforward neural network, W and U are S (t-1) and X t .

[0021] Preferably, the process of preprocessing the data in step S2 is as follows:

[0022] S21: Perform data cleaning on the collected time series work data, including duplicate data and invalid data:

[0023] S22: performing data conversion processing on the time series work data after the cleaning process;

[0024] S23: performing data normalization processing on the time series working data after the data conversion processing.

[0025] Preferably, the process of initializing weights and thresholds of the fault features of the extracted sequential working data of each component in step S4 is to assign preset initial weights and thresholds to the fault features of each component.

[0026] Preferably, the specific process of optimizing the weight and threshold of the fault characteristics of the sequential working data of each component in step S5 is as follows:

[0027] S51: Set the learning rate, number of iterations and error accuracy of the model;

[0028] S52: Initialize the weight and threshold of each fault feature;

[0029] S53: Calculate the input and output of the hidden layer at the current time step;

[0030] S54: Calculate the input and output of the output layer of the current time step;

[0031] S55: Calculate the global error of the current time step;

[0032] S56: Based on the next time step, execute step S53 to determine whether the global error is less than the error precision. If so, output the current weight and threshold. If not, update the current weight and threshold and re-execute step S53.

[0033] Preferably, the server hardware fault prediction result in step S6 includes normal server hardware and server hardware fault, and the data of server hardware fault includes the severity corresponding to the fault category, and the severity includes light, medium and severe.

[0034] Preferably, after step S5, the predictability of the server fault prediction model is also evaluated, and the evaluation indicators include fault detection rate, false alarm rate, false alarm rate and missed alarm rate;

[0035] The fault detection rate refers to the ratio of the number of correct predictions in a certain category to the total number of the same category; the false alarm rate refers to the ratio of the number of a certain category of faults predicted to be other faults; the false alarm rate refers to the percentage of normal samples diagnosed as a certain type of fault; and the missed alarm rate refers to the percentage of a certain type of fault diagnosed as no fault.

[0036] In a second aspect, a server hardware failure early warning system based on a recurrent neural network is provided, which is used to implement any one of the server hardware failure early warning methods based on a recurrent neural network, including a data acquisition module, a model building module, a model optimization module and an early warning module;

[0037] The data acquisition module is used to collect the time-series working data of each key component of the server;

[0038] The model building module is used to build a server hardware failure prediction model;

[0039] The model optimization module is used to optimize various parameters of the constructed server hardware failure prediction model;

[0040] The early warning module is used to perform early warning processing when a hardware failure occurs based on the server hardware failure prediction result output by the server hardware failure prediction model;

[0041] The server hardware failure prediction model constructed is a recurrent neural network model.

[0042] Preferably, the warning information of the warning module includes the fault type, the severity of the fault and corresponding fault handling suggestions.

[0043] The beneficial effects of the present invention include:

[0044] The server hardware fault early warning system and method based on recurrent neural network provided by the present invention construct a server hardware fault early warning model; collect the sequential working data of each key component of the server in real time, perform data preprocessing on the collected sequential working data, and input it into the server hardware fault early warning model for training; extract the fault characteristics of the sequential working data of each component; initialize the weights and thresholds of the fault characteristics of the sequential working data of each component extracted by the server hardware fault early warning model; optimize the weights and thresholds of the fault characteristics of the sequential working data of each component; input the real-time working data into the server hardware fault early warning model, and output the server hardware fault prediction results in real time.

[0045] It provides more accurate server fault prediction capabilities, and can obtain hardware fault results more timely. It has dynamic adaptability and can dynamically adjust relevant parameters, which enhances flexibility and practicality. Since the system can collect, process and analyze the time series data of server working parameters in real time, and make predictions through intelligent algorithms, it can provide more accurate and timely early warning notifications, effectively reducing the false alarm rate and missed alarm rate. The prediction results include the fault type, the severity of the fault, and the corresponding fault handling suggestions, which help administrators quickly respond to and handle faults, and improve the reliability and stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The figure is a flow chart of the software fault pre-diagnosis method applicable to domestic servers of the present invention. DETAILED DESCRIPTION

[0047] The following is combined with Figure 1 The present invention is further described in detail:

[0048] See also Figure 1 , a server hardware failure early warning method based on recurrent neural network includes the following steps:

[0049] S1: Build a server hardware failure warning model;

[0050] S2: collect the time series work data of each key component of the server in real time, pre-process the collected time series work data, and input it into the server hardware fault warning model for training;

[0051] S3: Extract the fault characteristics of the time series working data of each component through the server hardware fault early warning model;

[0052] S4: Initialize weights and thresholds for the fault features of the time series working data of each component extracted by the server hardware fault early warning model;

[0053] S5: Optimizing the weights and thresholds of the fault characteristics of the sequential working data of each component;

[0054] S6: Input real-time working data into the server hardware failure warning model, and output the server hardware failure prediction results in real time.

[0055] By creating a server hardware fault prediction model, the model is a recurrent neural network model. Since the server's operating data belongs to time series working data, the recurrent neural network model takes time series data as input, recurses in the direction of sequence evolution, and all nodes are recursively linked in a chain. By collecting the time series working data of each key component of the server and preprocessing the collected data, the preprocessed data is input into the server hardware fault prediction model to train the server hardware fault prediction model so that it has the learning ability to meet the corresponding requirements. By extracting the fault features of the time series working data of each component and assigning initial weights and thresholds to each fault feature, and then optimizing the weights and thresholds of the fault features of the time series working data of each component, the prediction ability of the model is improved. Finally, based on the real-time time series working data of each component of the server as input, the prediction results of the server hardware fault are output.

[0056] Through the above process, the fault prediction system has more accurate server fault prediction capabilities, can obtain hardware fault results more timely, has dynamic adaptability, can dynamically adjust relevant parameters, and enhances flexibility and practicality. Since the system can collect, process and analyze the time series data of server working parameters in real time, and make predictions through intelligent algorithms, it can provide more accurate and timely early warning notifications, effectively reducing the false alarm rate and missed alarm rate. The prediction results include the fault type, the severity of the fault, and the corresponding fault handling suggestions, which help administrators quickly respond to and handle faults and improve the reliability and stability of the system.

[0057] The server hardware failure warning model constructed in step S1 includes an input layer, a hidden layer, and an output layer, and its given sequence input learning data is:

[0058] X={X1+X2+···+X t};

[0059] Among them, X1, X2, X t is the time series data, t is the expansion length of the input learning data;

[0060] The formula for a cycle unit at time t is:

[0061] S t =f(W*S (t-1) , U*X t );

[0062] Among them, S t is the hidden layer, t is the cycle unit time; S (t-1) The current time S t The hidden layer of the previous time step, S t Need to use the S of the previous time step (t-1) , f is the encapsulated feedforward neural network, W and U are S (t-1) and X t .

[0063] In another embodiment, the server hardware fault model can also use a long short-term memory unit, which includes three gates, an input gate, a forget gate, and an output gate, and the three gates establish a self-loop for the internal state of the long short-term memory unit. The input gate determines the update of the internal state by the input of the current time step and the system state of the previous time step, the forget gate determines the update of the internal state of the previous time step to the internal state of the current time step, and the output gate determines the update of the internal state to the system state.

[0064] The process of preprocessing the data in step S2 is as follows:

[0065] S21: Perform data cleaning on the collected time series work data, including duplicate data and invalid data:

[0066] S22: performing data conversion processing on the time series work data after the cleaning process;

[0067] S23: performing data normalization processing on the time series working data after the data conversion processing.

[0068] The process of initializing the weights and thresholds of the fault features of the extracted sequential working data of each component in step S4 is to assign preset initial weights and thresholds to the fault features of each component.

[0069] The specific process of step S5 for optimizing the weight and threshold of the fault characteristics of the sequential working data of each component is as follows:

[0070] S51: Set the learning rate, number of iterations and error accuracy of the model;

[0071] S52: Initialize the weight and threshold of each fault feature;

[0072] S53: Calculate the input and output of the hidden layer at the current time step;

[0073] S54: Calculate the input and output of the output layer of the current time step;

[0074] S55: Calculate the global error of the current time step;

[0075] S56: Based on the next time step, execute step S53 to determine whether the global error is less than the error precision. If so, output the current weight and threshold. If not, update the current weight and threshold and re-execute step S53.

[0076] The server hardware fault prediction results in step S6 include normal server hardware and server hardware fault. The server hardware fault data include the severity and fault handling suggestions corresponding to the fault category. The severity includes light, medium and severe, and is reported to the early warning module for early warning processing. The administrator can obtain the fault type and severity of the components in the server hardware and the fault handling suggestions based on the early warning information of the early warning module, and can quickly respond to and handle the fault, thereby improving the reliability and stability of the system.

[0077] After step S5, the predictability of the server fault prediction model must be evaluated, and the evaluation indicators include fault detection rate, false alarm rate, false alarm rate and missed alarm rate. The fault detection rate refers to the ratio of the number of correct predictions in a certain category to the total number of the same category; the false alarm rate refers to the ratio of a certain category of faults predicted to be other faults; the false alarm rate refers to the percentage of normal samples diagnosed as a certain type of fault; and the missed alarm rate refers to the percentage of a certain type of fault diagnosed as no fault.

[0078] A server hardware fault early warning system based on a recurrent neural network is used to implement any one of the server hardware fault early warning methods based on a recurrent neural network, including a data acquisition module, a model construction module, a model optimization module and an early warning module. The data acquisition module is used to collect the time series working data of each key component of the server. The model construction module is used to build a server hardware fault prediction model. The model optimization module is used to optimize the various parameters of the constructed server hardware fault prediction model. The early warning module is used to perform early warning processing when a hardware fault exists based on the server hardware fault prediction result output by the server hardware fault prediction model; the constructed server hardware fault prediction model is a recurrent neural network model. The early warning information of the early warning module includes the fault type, the severity of the fault and the corresponding fault handling suggestions.

[0079] In summary, a server hardware fault warning model is constructed; the sequential working data of each key component of the server is collected in real time, and the collected sequential working data is preprocessed and input into the server hardware fault warning model for training; the fault characteristics of the sequential working data of each component are extracted; the weights and thresholds of the fault characteristics of the sequential working data of each component extracted by the server hardware fault warning model are initialized; the weights and thresholds of the fault characteristics of the sequential working data of each component are optimized; the real-time working data is input into the server hardware fault warning model, and the server hardware fault prediction results are output in real time.

[0080] It provides more accurate server fault prediction capabilities, and can obtain hardware fault results more timely. It has dynamic adaptability and can dynamically adjust relevant parameters, which enhances flexibility and practicality. Since the system can collect, process and analyze the time series data of server working parameters in real time, and make predictions through intelligent algorithms, it can provide more accurate and timely early warning notifications, effectively reducing the false alarm rate and missed alarm rate. The prediction results include the fault type, the severity of the fault, and the corresponding fault handling suggestions, which help administrators quickly respond to and handle faults, and improve the reliability and stability of the system.

[0081] The above-mentioned embodiments only express the specific implementation method of the present application, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the protection scope of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the technical solution concept of the present application, and these all belong to the protection scope of the present application.

Claims

1. A server hardware failure early warning method based on a recurrent neural network, characterized in that: The following steps are involved: S1: Build a server hardware failure warning model; S2: collect the time series work data of each key component of the server in real time, pre-process the collected time series work data, and input it into the server hardware fault warning model for training; S3: Extract the fault characteristics of the time series working data of each component through the server hardware fault early warning model; S4: Initialize weights and thresholds for the fault features of the time series working data of each component extracted by the server hardware fault early warning model; S5: Optimizing the weights and thresholds of the fault characteristics of the sequential working data of each component; S6: Input real-time working data into the server hardware failure warning model, and output the server hardware failure prediction results in real time.

2. The server hardware failure early warning method based on recurrent neural network according to claim 1 is characterized in that: The server hardware failure warning model constructed in step S1 includes an input layer, a hidden layer, and an output layer, and its given sequence input learning data is: X={X1+X 2+ …+X t }; Among them, X1, X2…X t is the time series data, t is the expansion length of the input learning data; The formula for a cycle unit at time t is: S t =f(W*S (t-1) ,U*X t ); Among them, S t is the hidden layer, t is the cycle unit time; S (t-1) Current time S t The hidden layer of the previous time step, S t Need to use the S of the previous time step (t-1) , f is the encapsulated feedforward neural network, W and U are S (t-1) and X t .

3. The server hardware failure early warning method based on recurrent neural network according to claim 1 is characterized in that: The process of preprocessing the data in step S2 is as follows: S21: Perform data cleaning on the collected time series work data, including duplicate data and invalid data: S22: performing data conversion processing on the time series work data after the cleaning process; S23: performing data normalization processing on the time series working data after the data conversion processing.

4. The server hardware failure early warning method based on recurrent neural network according to claim 1 is characterized in that: The process of initializing the weights and thresholds of the fault features of the extracted sequential working data of each component in step S4 is to assign preset initial weights and thresholds to the fault features of each component.

5. The server hardware failure early warning method based on recurrent neural network according to claim 4 is characterized in that: The specific process of step S5 for optimizing the weight and threshold of the fault characteristics of the sequential working data of each component is as follows: S51: Set the learning rate, number of iterations and error accuracy of the model; S52: Initialize the weight and threshold of each fault feature; S53: Calculate the input and output of the hidden layer at the current time step; S54: Calculate the input and output of the output layer of the current time step; S55: Calculate the global error of the current time step; S56: Based on the next time step, execute step S53 to determine whether the global error is less than the error precision. If so, output the current weight and threshold. If not, update the current weight and threshold and re-execute step S53.

6. The server hardware failure early warning method based on recurrent neural network according to claim 1 is characterized in that: The server hardware fault prediction result in step S6 includes normal server hardware and server hardware fault. The data of server hardware fault includes the severity corresponding to the fault category, and the severity includes light, medium and severe.

7. The server hardware failure early warning method based on recurrent neural network according to claim 1 is characterized in that: After step S5, the predictability of the server fault prediction model is also evaluated, and the evaluation indicators include fault detection rate, false alarm rate, false alarm rate and missed alarm rate; The fault detection rate refers to the ratio of the number of correct predictions in a certain category to the total number of the same category; The false alarm rate indicates the proportion of a certain type of fault that is predicted to be other faults; the false alarm rate refers to the percentage of normal samples diagnosed as a certain type of fault; the missed alarm rate refers to the percentage of a certain type of fault that is diagnosed as no fault.

8. A server hardware failure early warning system based on a recurrent neural network, used to implement the server hardware failure early warning method based on a recurrent neural network according to any one of claims 1 to 7, characterized in that: It includes data collection module, model building module, model optimization module and early warning module; The data acquisition module is used to collect the time-series working data of each key component of the server; The model building module is used to build a server hardware failure prediction model; The model optimization module is used to optimize various parameters of the constructed server hardware failure prediction model; The early warning module is used to perform early warning processing when a hardware failure occurs based on the server hardware failure prediction result output by the server hardware failure prediction model; The server hardware failure prediction model constructed is a recurrent neural network model.

9. The server hardware failure early warning system based on recurrent neural network according to claim 8 is characterized in that: The warning information of the warning module includes the fault type, the severity of the fault and the corresponding fault handling suggestions.

Images